Vise adds the workflow layer that turns your agent's context into a shippable integration. The agent still writes the code. Vise gives both the agent and the reviewer a better operating system to work in.
Guardrails on the AI coding loop. Vise gives the agent one defined workflow, runs 400+ platform-specific checks against the output, and blocks the merge until the result is green or attested.
Vise clarifies every ask against your actual app: existing screens, nav, and user context. Communities derive from the product. Feeds surface where users already look. Not a bolted-on "social demo" tab.
Build success is not runtime proof. Vise separates the two, requests real screenshots when runtime is claimed, and ships a versioned sp-vise/ contract your reviewer can gate merges on.
A local CLI and installable coding-agent skill for building social.plus SDK integrations with AI. Vise grounds the agent, runs 400+ platform-specific checks, and writes a versioned sp-vise/ contract with attested evidence for your reviewer.
No. Vise runs on your machine. It fetches public social.plus documentation and SDK version information, but it never uploads your source code, file contents, or queries. The contract and any evidence you record live in your repository.
MCP is the reference layer. Vise is the workflow layer. Use MCP when you want your AI assistant to find and read docs. Use Vise when the assistant is editing your app and you want a governed loop, deterministic checks, and a reviewer-ready contract at the end. Vise also runs as a local MCP server, so hosts can call its tools directly.
Claude Code, Cursor, VS Code with GitHub Copilot, OpenAI Codex, and any MCP-compatible host. Install the skill with vise install-skill --target claude (or your tool of choice).
Yes. Run vise baseline to snapshot pre-existing findings, then gate new work with vise check --new-only. Legacy findings stay visible without blocking the current change. The requested feature's completeness is never baselined.
Nothing you rely on today. Vise complements code review; it does not replace human judgment. The agent still writes the code, the check layer surfaces what compilers and linters miss, and the reviewer still ships.